fix: Handle API overload errors gracefully

- Detect Claude API 500 Overloaded errors
- Continue processing other files on error instead of aborting
- Log errors to stderr for visibility

This allows commits to succeed even if some AI requests fail due to rate limiting.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

automation/patcher.py

@@ -232,6 +232,9 @@ def call_model(model: ModelConfig, prompt: str, cwd: Path) -> str:
     # Check if we got output even if returncode is non-zero
     # (claude CLI returns 1 even on successful prompt responses)
     if result.stdout.strip():
+        # Check for API errors in the output
+        if "API Error:" in result.stdout and "Overloaded" in result.stdout:
+            raise PatchGenerationError("Claude API is overloaded (500 error) - please retry later")
         return result.stdout
 
     # Only raise error if we got nothing useful

automation/runner.py

@@ -72,14 +72,19 @@ def process(repo_root: Path, rules: RulesConfig, model: ModelConfig) -> int:
 
             final_instruction = merge_instructions(source_instruction, instruction, append)
 
-            generate_output(
-                repo_root=repo_root,
-                rules=rules,
-                model=model,
-                source_rel=src_rel,
-                output_rel=output_rel,
-                instruction=final_instruction,
-            )
+            try:
+                generate_output(
+                    repo_root=repo_root,
+                    rules=rules,
+                    model=model,
+                    source_rel=src_rel,
+                    output_rel=output_rel,
+                    instruction=final_instruction,
+                )
+            except Exception as e:
+                # Log error but continue processing other files
+                print(f"[runner] error generating {output_rel}: {e}", file=sys.stderr)
+                continue
 
     return 0